GPU Optimized Single Gaussian Skin Likelihood Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional pixel-based skin detection techniques face challenges in accuracy due to variations in lighting conditions, ethnicity, and unconstrained environments, with offline-trained classifiers prone to errors and limited adaptability.
Innovation Solution
The implementation of a single Gaussian based skin likelihood estimation method, which uses a GPU optimization technique to partition skin likelihood tasks between the CPU and GPU, improving the smoothness of skin maps and reducing computational costs by updating model parameters every 10-20 frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional pixel-based skin detection techniques are used, then skin detection can be performed, but accuracy deteriorates due to variations in lighting conditions, ethnicity, and unconstrained environments
Solution Approach 1:
The patent transforms the skin detection problem from pixel-level classification to superpixel-level classification, changing the fundamental parameter of analysis. By grouping pixels into superpixels and using color distribution parameters (mean and standard deviation in L*a*b* color space) rather than individual pixel values, the system achieves better adaptability to varying lighting conditions and ethnicities while improving detection accuracy
2Ease of operation
If offline-trained classifiers are used for skin detection, then classification can be performed, but reliability deteriorates due to errors and limited adaptability
Solution Approach 1:
The patent implements a dynamic skin detection system that adapts to different lighting conditions and ethnicities by using color distribution statistics (mean and standard deviation) of superpixels. Rather than relying on fixed offline-trained classifiers, the system dynamically adjusts its classification thresholds based on the actual color distribution in the input image, significantly improving reliability and robustness
Solution Approach 2:
The system incorporates feedback mechanisms by using the computed color distribution parameters of superpixels to dynamically adjust classification decisions. The mean and standard deviation values feed back into the classification process, allowing the system to adapt to varying conditions and improve reliability
3Measurement precision
If detailed pixel-level analysis is performed, then classification accuracy may improve, but computational cost increases
Solution Approach 1:
The patent applies segmentation by grouping pixels into superpixels, reducing the number of classification units from individual pixels to superpixel regions. This segmentation approach maintains classification accuracy by preserving the color distribution characteristics within each superpixel while dramatically reducing computational cost by analyzing fewer, larger units instead of every individual pixel
Data Source
AI summary
A system for performing single Gaussian skin detection is described herein. The system includes a memory and a processor. The memory is configured to receive image data. The processor is coupled to the memory. The processor is to generate a single Gaussian skin model based on a skin dominant region associated with the image data and a single Gaussian non-skin model based on a second region associated with the image data and to classify individual pixels associated with the image data via a discriminative skin likelihood function based on the single Gaussian skin model and the single Gaussian non-skin model to generate skin label data associated with the image data.